Research map / Vision-language and multimodal / Human and embodied modeling
Boosted Attention: Leveraging Human Attention for Image Captioning
Shi Chen, Qi Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add stimulus-based attention (Boosted Attention) (ablation) | Soft-attention ResNet-101 captioner · MSCOCO Karpathy test | CIDEr: 1.062 -> 1.122 | Improved |
| add stimulus-based attention (Boosted Attention) (ablation) | Soft-attention ResNet-101 captioner · MSCOCO Karpathy test | BLEU-4: 0.335 -> 0.354 | Improved |
| add stimulus-based attention (Boosted Attention) (ablation) | Soft-attention ResNet-101 captioner · Flickr30K test | CIDEr: 0.523 -> 0.586 | Improved |
| add stimulus-based attention (Boosted Attention) (ablation) | Soft-attention ResNet-101 captioner · Flickr30K test | BLEU-4: 0.267 -> 0.274 | Improved |
| add stimulus-based attention (Boosted Attention) (ablation) | Soft-attention ResNet-101 captioner · Flickr30K test | METEOR: 0.197 -> 0.208 | Improved |
| add pretrained saliency weights to stimulus-based attention module (BAM* to BAM) (ablation) | Boosted Attention · MSCOCO Karpathy test | CIDEr: 1.111 -> 1.122 | Improved |
| add pretrained saliency weights to stimulus-based attention module (BAM* to BAM) (ablation) | Boosted Attention · Flickr30K test | CIDEr: 0.571 -> 0.586 | Improved |
| add pretrained saliency weights to stimulus-based attention module (BAM* to BAM) (ablation) | Boosted Attention · MSCOCO Karpathy test | METEOR: 0.262 -> 0.265 | Improved |
| add architecture with randomly initialized stimulus module (BAM*) over baseline (ablation) | Soft-attention ResNet-101 captioner · MSCOCO Karpathy test | CIDEr: 1.062 -> 1.111 | Improved |
| ensemble of 4 BAM models vs ensemble Att2all on online server | Boosted Attention · MSCOCO online test (C5) | CIDEr: 1.147 -> 1.083 | Hurt |
| use stimulus-based attention alone as captioning attention | Stimulus attention · Karpathy test attention maps | correlation with captioning attention: CC = 0.222 | No clear effect |
| compare top-down attention maps with stimulus-based maps | Boosted Attention · Karpathy test 5000 images | CC and Spearman: CC = -0.256, Spearman = -0.369 | Mixed |
Closest papers by meaning
- Human Attention in Image Captioning: Dataset and Analysis (ICCV 2019 · similarity 0.94)
- Text-Guided Attention Model for Image Captioning (AAAI 2017 · similarity 0.92)
- Paying Attention to Descriptions Generated by Image Captioning Models (ICCV 2017 · similarity 0.91)
- Image Captioning With Semantic Attention (CVPR 2016 · similarity 0.90)
- Visual Attention Never Fades: Selective Progressive Attention ReCalibration for Detailed Image Captioning in Multimodal Large Language Models (ICML 2025 · similarity 0.89)
- Patch Matters: Training-free Fine-grained Image Caption Enhancement via Local Perception (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
